[PATCH 20/22] clocksource/drivers/exynos_mct: Fix Kconfig and add COMPILE_TEST option

[Arnd Bergmann]

On Tuesday 03 November 2015 09:40:02 Daniel Lezcano wrote:
> On 11/03/2015 01:59 AM, Krzysztof Kozlowski wrote:
> > On 03.11.2015 09:30, Krzysztof Kozlowski wrote:
> >> On 02.11.2015 21:56, Daniel Lezcano wrote:
> >>> Let the platform's Kconfig to select the clock instead of having a reverse
> >>> dependency from the driver to the platform options.
> >>
> >> Selecting user-visible symbols is rather discouraged so why not
> >> something like this:
> >>
> >> -       def_bool y if ARCH_EXYNOS
> >> -       depends on !ARM64
> >> +       bool "Exynos multi core timer driver"
> >> +       depends on ARCH_EXYNOS || (COMPILE_TEST && ARM)
> >
> > Nope, that was wrong as we loose auto-select on Exynos. Instead:
> > -       def_bool y if ARCH_EXYNOS
> > -       depends on !ARM64
> > +       bool "Exynos multi core timer driver" if ARM
> > +       depends on ARCH_EXYNOS || COMPILE_TEST
> > +       default y if ARCH_EXYNOS
> >
> > This way we avoid select (which is a reverse dependency for the driver),
> > have it auto-selectable and compile tested on arm.
> 
> I think you misunderstood the patch I sent.
> 
> It does two things:
> 
> 1. Follow the thumb of rule of the current Kconfig format
> 
>     - The timer driver is selected by the platform (exynos in this case)
>     - User can't select the driver in the menuconfig
>     - There is no dependency on the platform except for compilation test
> 
> 2. Add the COMPILE_TEST
> 
>     - User can select the driver for compilation testing. This is for 
> allyesconfig when doing compilation test coverage (exynos timer could be 
> compiled on other platform). As the delay code is not portable, we have 
> to restrict the compilation on the ARM platform, this is why there is 
> the dependency on ARM.
> 
> I am currently looking at splitting the delay code in order to prevent 
> this restriction on this driver and some others drivers.

I suspect this will come up again in the future. The problem is
really that drivers/clocksource has different rules from almost
everything else, by requiring the platform to 'select' the driver.

The second version that Krzysztof posted is how we handle this in
other driver subsystems, and I would generally prefer it to do this
consistently for everything, but John Stultz has in the past argued
strongly for using 'select' in all clocksource drivers. The reason
is that for each platform we know in advance which driver we want,
and there is never a need for the user to have to select the right
one.

	Arnd